Paleocystis globosa red tide automatic detection method based on video monitoring

Abstract

The invention discloses a method for automatically detecting red tide of Paleocystis globosa based on video monitoring, which comprises the following steps of: S1, acquiring image data of a sea surface by using a high-definition monitoring camera on a buoy platform, transmitting the image data to a database in real time, and preprocessing the image data; s2, making a labeling scheme, labeling the preprocessed image data, and constructing a training sample library; s3, training and optimizing the YOLOv8 target detection model to obtain a globocystis globosa recognition model; s4, carrying out frame extraction processing on a video monitoring picture to be identified according to a certain time interval, inputting each frame of picture into the globocystis globocystis identification model for identification, and automatically generating globocystis globocystis red tide monitoring information; according to the method, a video monitoring technology and an artificial intelligence target detection algorithm are combined, so that efficient and accurate monitoring of capsules generated during red tide of the globocystis globocystis can be realized, and the real-time performance and the automation level of red tide monitoring of the globocystis globocystis can be remarkably improved.

Description

Technical Field

[0001] The invention belongs to the field of computer technology, and in particular relates to an automatic detection method for red tide of Phaeocystis sphericalensis based on video monitoring. Background Art

[0002] Pseudocystis spherica is widely distributed in the world's oceans and often causes large-scale red tides. Pseudocystis spherica red tides not only lead to water quality deterioration and fish deaths, but may also block the cooling water systems of coastal nuclear power plants, threatening the normal operation and safety of nuclear power. In addition, in the late stage of the red tide, the cysts of Pseudocystis spherica will rupture and decompose, releasing a large amount of polysaccharides into the seawater, causing a large amount of foamy organic matter to float on the sea surface, further polluting the marine and coastal ecological environment and affecting the coastal landscape. Therefore, the Pseudocystis spherica red tide has attracted great attention from governments and the public at all levels.

[0003] When a red tide of P. sphericalis occurs, centimeter-sized "giant cysts" are usually formed. These cysts are visible to the naked eye and can be monitored in real time through video surveillance. Compared with traditional manual sampling, satellite or drone monitoring methods, video surveillance technology has several unique advantages. First, video surveillance can provide high-frequency and continuous observation data to track the dynamic changes of red tides in real time, which is crucial for timely discovering the development trend of red tides and assessing their degree of harm. Secondly, the video surveillance system installed on the buoy does not need to rely on complex external equipment and can operate stably under various sea conditions, which is particularly suitable for long-term monitoring. In addition, video data can be directly used for image analysis, and combined with artificial intelligence technology, it can further improve the monitoring accuracy and automation level, thereby achieving efficient and relatively low-cost red tide monitoring. Therefore, as an emerging monitoring method, video surveillance technology provides a convenient, continuous and efficient solution for real-time monitoring of P. sphericalis red tides. However, there is currently no technical solution for automatic detection of P. sphericalis red tides based on image data collected by video surveillance at home and abroad. Therefore, in order to meet the needs of emergency monitoring of red tides, it is crucial to construct an automatic detection method for red tides suitable for video surveillance, a high-frequency sampling method. Summary of the invention

[0004] To solve the above problems, the present invention proposes an automatic detection method for spherical brown algae red tide based on video surveillance. The method combines video surveillance technology with an artificial intelligence target detection algorithm, and through re-labeling and transfer learning after introducing different scenes, it is found that the recognition effect is significantly improved. The cysts produced during the spherical brown algae red tide can be efficiently and accurately monitored, thereby significantly improving the real-time and automation level of spherical brown algae red tide monitoring, which can not only provide a scientific basis for the early warning and management of red tides, but also provide favorable support for marine environmental protection and red tide prevention and control.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring comprises the following steps:

[0007] S1. Use the high-definition monitoring camera on the buoy platform to collect image data of the sea surface, transmit the image data to the database in real time, and pre-process the image data;

[0008] S2. Develop a labeling scheme, label the preprocessed image data, and build a training sample library;

[0009] S3, training and optimizing the YOLOv8 target detection model to obtain a spherical brown algae recognition model;

[0010] S4, extracting frames of the video surveillance images to be identified at a certain time interval, and then inputting each frame of the image into the spherical brown algae recognition model for identification, and automatically generating spherical brown algae red tide monitoring information.

[0011] Preferably, the preprocessing in step S1 includes denoising, contrast enhancement and image transformation augmentation, which are used to remove irrelevant information in the image, restore and enhance useful information, and enhance the detectability of key features.

[0012] Preferably, the specific process of denoising in step S1 is: using a median filtering algorithm, for each image, first determine a filtering window, then traverse each pixel in the image, sort the pixel values ​​in the filtering window, and take the median as the new value of the current pixel.

[0013] Preferably, the specific process of contrast enhancement in step S1 is: using a histogram equalization method to enhance the image contrast, first counting the frequency of occurrence of each gray level in the image, constructing a gray histogram, and then calculating a gray transformation function based on the total number of pixels in the image and the frequency of each gray level. Through the gray transformation function, the gray values ​​of the darker or brighter areas in the original image are stretched to make the gray distribution of the image more uniform and improve the contrast of the image.

[0014] Preferably, the specific process of the image transformation and augmentation in step S1 is: first perform an image segmentation step to obtain the height and width of the original image, evenly divide the original image into 3×3 small pictures in the form of a nine-square grid, calculate the height and width of each small picture, and extract and store the small pictures in sequence from left to right and from top to bottom; then perform a small picture augmentation operation, randomly rotate each small picture, randomly select an angle from 0 degrees and 180 degrees, take the center of the small picture as the rotation point, use the image rotation algorithm to rotate the small picture around the center by a corresponding angle, and perform a random flipping operation at the same time, randomly decide whether to perform horizontal flipping and vertical flipping, and if it is decided to perform horizontal flipping The horizontal flip function is used to mirror the small image horizontally. If it is decided to flip vertically, the vertical flip function is used to mirror it vertically, thus obtaining small images with different rotation and flipping states; finally, image stitching is implemented to create a blank target image with 1080p resolution and the same number of color channels as the original image, and the pixels are initialized. Any 9 small images after random augmentation are stitched into the target image in a 3×3 layout in sequence. According to the size of the small image and the target position, the pixels of the small image are accurately copied to the corresponding area of ​​the target image with the help of image copying means, and finally a new image is formed after segmentation, augmentation and stitching.

[0015] Preferably, the labeling scheme in step S2 is to mark the position, length and width of the target in the picture, wherein the target includes the cyst of the spherical brown cyst; the specific process of labeling is: using the built-in EfficientSam function in the Labelme labeling software to semi-automatically label the cyst of the spherical brown cyst, and by clicking the cyst target, the pixel range contained in the cyst is wrapped in a dotted line manner; after completing the labeling of a series of targets, a labeling file of the corresponding image is formed; and then through batch processing, the labeling file is converted into a text file that can be read by the YOLOv8 target detection model.

[0016] Preferably, the specific process of training the YOLOv8 target detection model in step S3 is:

[0017] S31, using the annotated data in the training sample library as the input data set, and adjusting the image data to 448×448×3; the annotated data includes the category c of the target and the bounding box coordinates (x, y, w, h) of the target in the image, where x and y are the horizontal and vertical coordinates of the center point of the target, respectively; w and h are the width and height of the target, respectively;

[0018] The backbone network of the S32 and YOLOv8 target detection models is based on a convolutional neural network structure. It extracts features from the input image through a series of convolutional layers and downsampling layers. The calculation formula for the convolution operation is:

[0019] F i,j,k =∑m,n,c W m,n,c.k I i+m,j+n,c +b k ,

[0020] Among them, F i,j,k Represents the elements in the feature map; W m,n,c,k Represents the convolution kernel weight; I i+m,j+n,c is the pixel value of the input image; b k is the bias term; after the convolution operation, the spatial features of the image are extracted and downsampled;

[0021] The S33 and YOLOv8 target detection models use a feature pyramid network structure for multi-scale fusion, which can grasp global information and local details by fusing features of different scales.

[0022] S34. In the target detection stage, the YOLOv8 target detection model predicts the category and bounding box information of the target on each grid of the image. For each predicted grid, the YOLOv8 target detection model outputs a vector:

[0023] [x1,y1,w1,h1,p c ,p1,p2,p3,……,p k ],

[0024] Among them, x1 and y1 are the horizontal and vertical offsets of the center point of the predicted bounding box relative to the grid; w1 and h1 are the width and height of the predicted bounding box respectively; p c is the confidence of the target; p1, p2, p3, ..., p k are the probability distributions of each target category;

[0025] The S35 and YOLOv8 target detection models are trained using the following loss functions:

[0026] L=L bbox +L obj +L class ,

[0027] Among them, L is the loss function; L bbox L is the bounding box loss, which measures the difference between the predicted bounding box and the true bounding box and is calculated using IoU loss or GIoU loss; obj is the confidence loss, which measures the difference between the predicted target confidence and the actual target existence; L class The category loss measures the difference between the predicted target category and the true category, and uses cross entropy loss or focal loss to improve the recognition ability of a small number of categories.

[0028] Preferably, the optimization of the YOLOv8 target detection model in step S3 includes increasing the sample size and transfer learning;

[0029] The specific process of increasing the sample size is as follows: in the data collection stage, in addition to the initially collected and annotated data, new spherical brown cysts algae image data are continuously collected; wherein the new spherical brown cysts algae image data include spherical brown cysts algae images in different or special scenes; the newly collected spherical brown cysts algae image data are manually annotated according to the formulated annotation scheme; the annotated data are added to the original sample library to expand the sample library;

[0030] The specific process of the transfer learning is as follows: in the training stage of the YOLOv8 target detection model, the model weights of the recognition task are used as pre-training weights; the pre-training weights are input into a new model for recognizing spherical brown algae; fine-tuning is performed on a new dataset of spherical brown algae to improve the learning efficiency of the new task, and the knowledge accumulation of the existing model is utilized to reduce the training time and improve the adaptability and accuracy of the model.

[0031] Preferably, the spherical brown cysts recognition model described in step S4 automatically detects the spherical brown cysts, marks the specific position on the cyst image, and automatically counts the size and number of the spherical brown cysts recognized in each frame of the image; combined with the time of image acquisition, the time series data of the size and number of cysts within the monitoring range during the spherical brown cyst red tide is formed; for the entire red tide event, the time series data is used to automatically count the average value, maximum value, minimum value, standard deviation and total number of cyst size, form the time change trend of the statistical values ​​of each parameter, and generate statistical charts.

[0032] After adopting the above technical scheme, the present invention has the following beneficial effects: by combining video surveillance technology with artificial intelligence target detection algorithm, the present invention can quickly obtain monitoring pictures of spherical brown cysts in complex sea conditions, and can automatically identify the acquired pictures, and extract information such as the size and number of spherical brown cysts; especially after re-labeling and transfer learning after introducing different scenes, it is found that the recognition effect is significantly improved. For example, in the case of stacked multi-cysts, the stacked cysts can be effectively separated and identified one by one, such as Figure 4 and Figure 5 As shown; it can realize efficient and accurate monitoring of the cysts produced during the red tide of P. sphericalensis, and improve the time resolution and real-time performance of the red tide monitoring of P. sphericalensis, thereby significantly improving the real-time and automation level of the red tide monitoring of P. sphericalensis, which can not only provide a scientific basis for the early warning and management of red tides, but also provide favorable support for marine environmental protection and red tide prevention and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 It is a schematic diagram of the process of the present invention;

[0034] Figure 2 It is a flowchart of the present invention;

[0035] Figure 3 A schematic diagram of the network structure of the YOLOv8 target detection model of the present invention;

[0036] Figure 4 This is a diagram showing the recognition results of the spherical brown cysts when the YOLOv8 target detection model of the present invention is not optimized;

[0037] Figure 5 This is a diagram showing the recognition results of the spherical brown algae cysts after the YOLOv8 target detection model of the present invention is optimized. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] like Figures 1 to 5 As shown, a method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring comprises the following steps:

[0040] S1. Use the high-definition monitoring camera on the buoy platform to collect image data of the sea surface, transmit the image data to the database in real time, and pre-process the image data;

[0041] The preprocessing in step S1 includes denoising, contrast enhancement and image transformation augmentation, which are used to remove irrelevant information in the image, restore and enhance useful information, and enhance the detectability of key features;

[0042] The specific process of denoising in step S1 is: using the median filtering algorithm, for each image, first determine a filtering window, then traverse each pixel in the image, sort the pixel values ​​in the filtering window, and take the median value as the new value of the current pixel;

[0043] The specific process of contrast enhancement in step S1 is: using a histogram equalization method to enhance the image contrast, first counting the frequency of each gray level in the image, constructing a gray histogram, and then calculating a gray transformation function according to the total number of pixels in the image and the frequency of each gray level. Through the gray transformation function, the gray value of the darker or brighter area in the original image is stretched to make the gray distribution of the image more uniform and improve the contrast of the image;

[0044] The specific process of the image transformation and augmentation in step S1 is as follows: first, an image segmentation step is performed to obtain the height and width of the original image, and the original image is evenly divided into 3×3 small pictures in the form of a nine-square grid, the height and width of each small picture are calculated, and the small pictures are extracted and stored in sequence from left to right and from top to bottom; then, a small picture augmentation operation is performed, and each small picture is randomly rotated, an angle is randomly selected from 0 degrees and 180 degrees, and the center of the small picture is used as the rotation point. The image rotation algorithm is used to rotate the small picture around the center by a corresponding angle, and a random flipping operation is performed at the same time, and whether to perform horizontal flipping and vertical flipping is randomly determined. If it is decided to perform horizontal flipping, the small picture is rotated by the image rotation algorithm. Use the horizontal flip function to mirror the small image horizontally. If you decide to flip it vertically, use the vertical flip function to mirror it vertically, so as to obtain small images with different rotation and flip states. Finally, implement image stitching, create a blank target image with 1080p resolution and the same number of color channels as the original image, and initialize the pixels. Then, stitch any 9 small images after random augmentation into the target image in a 3×3 layout. According to the size of the small image and the target position, use image copying to accurately copy the pixels of the small image to the corresponding area of ​​the target image, and finally form a new image after segmentation, augmentation and stitching.

[0045] S2. Develop a labeling scheme, label the preprocessed image data, and build a training sample library;

[0046] The labeling scheme described in step S2 is to mark the position, length and width of the target in the picture, wherein the target includes the cyst of the spherical brown cyst; the specific process of labeling is: using the built-in EfficientSam function in the Labelme labeling software to semi-automatically label the cyst of the spherical brown cyst, and by clicking the cyst target, the pixel range contained in the cyst is wrapped in a dotted line manner; after completing the labeling of a series of targets, a labeling file of the corresponding image is formed; and then, through batch processing, the labeling file is converted into a text file that can be read by the YOLOv8 target detection model;

[0047] S3, training and optimizing the YOLOv8 target detection model to obtain a spherical brown algae recognition model;

[0048] The specific process of training the YOLOv8 target detection model in step S3 is:

[0049] S31, using the annotated data in the training sample library as the input data set, and adjusting the image data to 448×448×3; the annotated data includes the category c of the target and the bounding box coordinates (x, y, w, h) of the target in the image, where x and y are the horizontal and vertical coordinates of the center point of the target, respectively; w and h are the width and height of the target, respectively;

[0050] The backbone network of the S32 and YOLOv8 target detection models is based on a convolutional neural network structure. It extracts features from the input image through a series of convolutional layers and downsampling layers. The calculation formula for the convolution operation is:

[0051] F i,j,k =∑ m,n,c W m,n,c.k I i+m,j+n,c +b k ,

[0052] Among them, F i,j,k Represents the elements in the feature map; W m,n,c,k Represents the convolution kernel weight; I i+m,j+n,c is the pixel value of the input image; b k is the bias term; after the convolution operation, the spatial features of the image are extracted and downsampled;

[0053] The S33 and YOLOv8 target detection models use a feature pyramid network structure for multi-scale fusion, which can grasp global information and local details by fusing features of different scales.

[0054] S34. In the target detection stage, the YOLOv8 target detection model predicts the category and bounding box information of the target on each grid of the image. For each predicted grid, the YOLOv8 target detection model outputs a vector:

[0055] [x1,y1,w1,h1,p c ,p1,p2,p3,……,p k ],

[0056] Among them, x1 and y1 are the horizontal and vertical offsets of the center point of the predicted bounding box relative to the grid; w1 and h1 are the width and height of the predicted bounding box respectively; p c is the confidence of the target; p1, p2, p3, ..., p k are the probability distributions of each target category;

[0057] The S35 and YOLOv8 target detection models are trained using the following loss functions:

[0058] L=L bbox +L obj +L class ,

[0059] Among them, L is the loss function; L bbox L is the bounding box loss, which measures the difference between the predicted bounding box and the true bounding box and is calculated using IoU loss or GIoU loss; objis the confidence loss, which measures the difference between the predicted target confidence and the actual target existence; L class Category loss measures the difference between the predicted target category and the true category, and uses cross entropy loss or focal loss to improve the recognition ability of a small number of categories;

[0060] In step S3, the YOLOv8 target detection model is optimized including increasing the sample size and transfer learning;

[0061] The specific process of increasing the sample size is as follows: in the data collection stage, in addition to the initially collected and annotated data, new spherical brown cysts algae image data are continuously collected; wherein the new spherical brown cysts algae image data include spherical brown cysts algae images in different or special scenes; the newly collected spherical brown cysts algae image data are manually annotated according to the formulated annotation scheme; the annotated data are added to the original sample library to expand the sample library;

[0062] The specific process of the transfer learning is as follows: in the training stage of the YOLOv8 target detection model, the model weights of the recognition task are used as pre-training weights; the pre-training weights are input into the new model for the recognition of spherical brown algae; fine-tuning is performed on the new data set of spherical brown algae to improve the learning efficiency of the new task, and the knowledge accumulation of the existing model is used to reduce the training time and improve the adaptability and accuracy of the model;

[0063] S4, extracting frames of the video surveillance image to be identified at a certain time interval, and then inputting each frame of the image into the spherical brown cyst algae recognition model for identification, and automatically generating spherical brown cyst algae red tide monitoring information;

[0064] The spherical brown cysts recognition model described in step S4 automatically detects the spherical brown cysts, marks the specific position on the cyst image, and automatically counts the size and number of the spherical brown cysts recognized in each frame of the image; combined with the time of image acquisition, the time series data of the size and number of cysts within the monitoring range during the spherical brown cysts red tide is formed; for the entire red tide event, the time series data is used to automatically count the average value, maximum value, minimum value, standard deviation and total number of cyst sizes, form the time change trend of the statistical values ​​of each parameter, and generate statistical charts.

[0065] Performance Test:

[0066] The performance test was performed using the YOLOv8 target detection model before and after optimization, respectively. The YOLOv8 target detection model before optimization did not use the optimization method of increasing the sample size and transfer learning in step S3; the YOLOv8 target detection model after optimization used the optimization method of increasing the sample size and transfer learning in step S3, and the optimized YOLOv8 target detection model was the spherical brown cyst algae recognition model of the present application. The detection results of the YOLOv8 target detection model before and after optimization are shown in Table 1.

[0067] Table 1: Detection results of YOLOv8 target detection model before and after optimization

[0068] Model Evaluation Metrics Before optimization After optimization Optimization rate Precision P 0.908 0.960 ↑5.7％ Recall R 0.564 0.625 ↑10.8％ mAP50 0.675 0.721 ↑6.8％ mAP@[0.5:0.95] 0.481 0.674 ↑40.1％

[0069] From Table 1, Figure 4 and Figure 5 It can be seen that after the re-labeling and transfer learning after the introduction of different scenes, the recognition effect is found to be significantly improved. For example, in the case of stacked polycystic bodies, the stacked polycystic bodies can be effectively separated and identified one by one. Therefore, the present invention can effectively improve the accuracy and reliability of the spherical brown cystic algae recognition model in actual marine environment monitoring by combining the two optimization methods of increasing the sample size and transfer learning.

[0070] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring, characterized in that: The following steps are involved: S1. Use the high-definition monitoring camera on the buoy platform to collect image data of the sea surface, transmit the image data to the database in real time, and pre-process the image data; S2. Develop a labeling scheme, label the preprocessed image data, and build a training sample library; S3, training and optimizing the YOLOv8 target detection model to obtain a spherical brown algae recognition model; S4, extracting frames of the video surveillance images to be identified at a certain time interval, and then inputting each frame of the image into the spherical brown algae recognition model for identification, and automatically generating spherical brown algae red tide monitoring information.

2. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 1, characterized in that: The preprocessing in step S1 includes denoising, contrast enhancement and image transformation augmentation, which are used to remove irrelevant information in the image, restore and enhance useful information, and enhance the detectability of key features.

3. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 2, characterized in that: The specific process of denoising in step S1 is: using the median filtering algorithm, for each image, first determine a filtering window, then traverse each pixel in the image, sort the pixel values ​​in the filtering window, and take the median as the new value of the current pixel.

4. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 2, characterized in that: The specific process of contrast enhancement in step S1 is: using the histogram equalization method to enhance the image contrast, first counting the frequency of each gray level in the image, constructing a gray histogram, and then calculating a gray transformation function based on the total number of pixels in the image and the frequency of each gray level. Through the gray transformation function, the gray values ​​of the darker or brighter areas in the original image are stretched to make the gray distribution of the image more uniform and improve the contrast of the image.

5. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 2, characterized in that: The specific process of the image transformation and augmentation in step S1 is as follows: first, an image segmentation step is performed to obtain the height and width of the original image, and the original image is evenly divided into 3×3 small pictures in the form of a nine-square grid, the height and width of each small picture are calculated, and the small pictures are extracted and stored in sequence from left to right and from top to bottom; then, a small picture augmentation operation is performed, and each small picture is randomly rotated, an angle is randomly selected from 0 degrees and 180 degrees, and the center of the small picture is used as the rotation point. The image rotation algorithm is used to rotate the small picture around the center by a corresponding angle, and a random flipping operation is performed at the same time, and whether to perform horizontal flipping and vertical flipping is randomly determined. If it is decided to perform horizontal flipping, the small picture is rotated by the image rotation algorithm. The small image is horizontally mirrored using the horizontal flip function. If it is decided to flip vertically, the vertical flip function is used to perform a vertical mirror operation, thereby obtaining small images with different rotation and flipping states. Finally, image stitching is implemented to create a blank target image with a 1080p resolution and the same number of color channels as the original image, and the pixels are initialized. Any 9 small images that have been randomly augmented are sequentially stitched into the target image in a 3×3 layout. According to the size of the small image and the target position, the pixels of the small image are accurately copied to the corresponding area of ​​the target image with the help of image copying means, and finally a new image is formed after segmentation, augmentation and stitching.

6. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 1, characterized in that: The labeling scheme described in step S2 is to mark the position, length and width of the target in the picture, wherein the target includes the cyst of the spherical brown cyst; the specific process of labeling is: using the built-in EfficientSam function in the Labelme labeling software to semi-automatically label the cyst of the spherical brown cyst, and by clicking the cyst target, the pixel range contained in the cyst is wrapped in a dotted line manner; after completing the labeling of a series of targets, a labeling file of the corresponding image is formed; and then through batch processing, the labeling file is converted into a text file that can be read by the YOLOv8 target detection model.

7. The automatic detection method of Phaeocystis sphericalensis red tide based on video monitoring as claimed in claim 1, characterized in that: The specific process of training the YOLOv8 target detection model in step S3 is: S31, using the annotated data in the training sample library as the input data set, and adjusting the image data to 448×448×3; the annotated data includes the category c of the target and the bounding box coordinates (x, y, w, h) of the target in the image, where x and y are the horizontal and vertical coordinates of the center point of the target, respectively; w and h are the width and height of the target, respectively; The backbone network of the S32 and YOLOv8 target detection models is based on a convolutional neural network structure. It extracts features from the input image through a series of convolutional layers and downsampling layers. The calculation formula for the convolution operation is: F i，j，k =∑ m,n，c W m,n，c.k ·I i+m,j+n,c +b k , Among them, F i,j,k Represents the elements in the feature map; W m,n,c,k Represents the convolution kernel weight; I i+m,j+n,c is the pixel value of the input image; b k is the bias term; after the convolution operation, the spatial features of the image are extracted and downsampled; The S33 and YOLOv8 target detection models use a feature pyramid network structure for multi-scale fusion, which can grasp global information and local details by fusing features of different scales. S34. In the target detection stage, the YOLOv8 target detection model predicts the category and bounding box information of the target on each grid of the image. For each predicted grid, the YOLOv8 target detection model outputs a vector: [x1, y1, w1, h1, p c ,p1,p2,p3,……,p k ], Among them, x1 and y1 are the horizontal and vertical offsets of the center point of the predicted bounding box relative to the grid; w1 and h1 are the width and height of the predicted bounding box respectively; p c is the confidence of the target; p1, p2, p3, ..., p k are the probability distributions of each target category; The S35 and YOLOv8 target detection models are trained using the following loss functions: L=L bbox +L obj +L class , Among them, L is the loss function; L bbox L is the bounding box loss, which measures the difference between the predicted bounding box and the true bounding box and is calculated using IoU loss or GIoU loss; obj is the confidence loss, which measures the difference between the predicted target confidence and the actual target existence; L class The category loss measures the difference between the predicted target category and the true category, and uses cross entropy loss or focal loss to improve the recognition ability of a small number of categories.

8. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 1, characterized in that: In step S3, YOLO v 8. Optimization of target detection models includes increasing sample size and transfer learning; The specific process of increasing the sample size is as follows: in the data collection stage, in addition to the initially collected and annotated data, new spherical brown cysts algae image data are continuously collected; wherein the new spherical brown cysts algae image data include spherical brown cysts algae images in different or special scenes; the newly collected spherical brown cysts algae image data are manually annotated according to the formulated annotation scheme; the annotated data are added to the original sample library to expand the sample library; The specific process of transfer learning is as follows: v 8. During the training phase of the target detection model, the model weights of the recognition task are used as pre-training weights. The pre-training weights are input into a new model for identifying P. sphericalis. Fine-tuning is performed on a new dataset of P. sphericalis to improve the learning efficiency of the new task, and the knowledge accumulation of the existing model is utilized to reduce the training time and improve the adaptability and accuracy of the model.

9. The method for automatically detecting red tide of Phaeocystis sphericalensis based on video monitoring as claimed in claim 1, characterized in that: The spherical brown cysts recognition model described in step S4 automatically detects the spherical brown cysts, marks the specific position on the cyst image, and automatically counts the size and number of the spherical brown cysts recognized in each frame of the image; combined with the time of image acquisition, the time series data of the size and number of cysts within the monitoring range during the spherical brown cysts red tide is formed; for the entire red tide event, the time series data is used to automatically count the average value, maximum value, minimum value, standard deviation and total number of cyst sizes, form the time change trend of the statistical values ​​of each parameter, and generate statistical charts.

Images